DE69320746T2 - Raster image improvement using a reduced template memory - Google Patents

Description

FIELD OF INVENTION

This invention relates to a pixel image enhancement technique and more particularly to a method and apparatus for enhancing pixel images using pre-stored pixel image models.

BACKGROUND OF THE INVENTION

A variety of printers are designed to produce a printed image by printing a series of picture elements (pixels) on a printing medium such as paper. Laser printers, which are a form of electrophotographic printer, produce an image by scanning a laser beam across a charged surface of a photosensitive drum in a series of raster scan lines. Each scan line is divided into pixel cells, with the laser beam being modulated so that pixel cells are either exposed or not, depending on the particular image being printed. Whenever a pixel cell is illuminated by the laser beam, the photosensitive drum is discharged and can then be toned in the known manner.

In laser printers, a bit map memory typically comprises a digitized quantized image of a desired analog image. As such, it consists of a large number of discrete pixels organized in a predetermined raster pattern. The final resolution of an image produced by a laser printer depends on the number of pixels printed per inch. However, even at a high pixel count resolution level, diagonal lines and boundaries between different regions of an image appear as a toothed step or stair distortion that is visible to the human eye.

Prior art documents have considered methods for improving pixel image resolution. In U.S. Patent 3,573,789, issued to Sharp, a 3 x 3 pixel cell subset is sequentially stepped across a bit map, with, at each step, a center pixel being modified according to the logic states of each of the eight surrounding pixels. The center pixel can be printed as either a full pixel or as a quarter, half, or three-quarter pixel.

U.S. Patent 4,437,122, issued to Walsh et al., describes a similar nine-pixel sub-image, but enhances the center pixel by selectively energizing one or more of the nine sub-pixels contained in the center pixel. A table lookup is performed to match a pre-stored pixel image with an input pixel image. Among the pre-stored pixel images are rotated versions (90º, 180º, and 270º) of the elementary pre-stored pixel image. If a match is found, it is known that the center pixel must be modified by turning on (or off) one or more of the nine sub-pixels in the center pixel.

U.S. Patent 4,450,483, issued to Coviello, performs a similar pixel subimage matching test as taught by Walsh et al., but according to Coviello, only decides whether the entire central pixel is to be modified, not substituting a set of subpixels for it.

US Patent 4,847,641, issued to Tung, the assignee of the present application, also improves the printing of bitmap images by matching pixel subimages with predetermined stored models. len or patterns to detect the occurrence of preselected bit map features. Whenever a match occurs, a signal is generated to produce a corrected pixel that replaces a center pixel in the pixel sub-image. The stored pixel models represent pixel configurations that are common to all bit map images and are compiled into a match check table.

Figure 1 shows the logic used by Tung to obtain an output subpixel. An input pixel window 10, which has a matrix of cells (the edges of which are removed), is accessed from a pixel bitmap. The center pixel 12 is the pixel that is changed (if necessary) based on a comparison of the input window 10 with a plurality of models stored in a model memory 14. A circle in a pixel cell represents a white pixel, a black circle represents a black pixel, and any cell that has neither a black nor a white circle is a "don't care" cell.

In Tung's system, the input window 10 is passed to a model memory 14 and compared with a plurality of models. The models are stored in a non-rotated version as well as in versions of 90º, 180º and 270º rotation. Each stored model is associated with a subpixel control indication which causes a laser printer to produce one of the subpixels 16, 18, 20 or 22. In the case shown in Figure 1, the input window 10 matches the model 24, resulting in the output lines 26, 28 and 30 being in the OFF state while the output line 32 is in the ON state. The ON state of the output line 32 signals a laser controller in the laser printer to print the output subpixel 34. More specifically, the output line 32 causes the laser printer to substitute the central pixel 12 with the output subpixel 34 to provide improved edge representation The pixel output signals on lines 26, 28, 30 and 32 are shown as L50 (ie the left 50% of the pixel is ON), R50 (ie the right half of the pixel is ON), T50 (ie the top of the pixel is ON) and B50 (ie the bottom of the pixel is ON).

In performing Tung's method, hundreds of models are envisaged. Many of these models are rotated and/or mirrored versions of the elementary models. Consequently, considerable redundant information exists in the rotated models in the Tung memory. However, such information is required to capture all orientations of the particular input image distortion that each elementary model is designed to capture. Similar methods involving large numbers of models are also disclosed in US-A-4847641 and EP-A-0549314.

Accordingly, it is an object of this invention to provide an improved pixel resolution augmentation method that uses a reduced model memory.

It is a further object of this invention to provide an improved pixel image display resolution enhancement system that is fast in execution and uses a minimal number of pre-stored comparison models.

SUMMARY OF THE INVENTION

According to one aspect, the invention provides a method for improving the reproduction of pixel images, comprising the steps of: storing a raster scan bit map of an image; selecting an input window from the bit map image comprising a plurality of row segments of bits representing pixels, the input window having a center pixel bit; comparing a Subset of the bits in the input window (including the center pixel bit) with a plurality of prediction bit subsets, each prediction subset represented in a plurality of rotational orientations, and upon finding a match with one of the prediction bit subsets, rotating the input window at the angle and direction indicated by a command associated with a matching prediction subset (if one exists); comparing the rotated input window with a limited set of sample windows to determine a match of certain bit states in the rotated input window with certain bits of a stored sample window; and upon determining that such a match exists, substituting the center pixel bit of the input window with an improved pixel representation.

According to a further aspect, the invention provides apparatus for improving the reproduction of images represented by a plurality of pixels. The apparatus comprises storage means for storing successive lines of a bit map image and means for selecting a sampling window of pixels from the bit map image in the storage means, the sampling window of pixels comprising a plurality of line segments of bits representing the pixels, the sampling window having a center bit representing a center pixel in the sampling window. Predicting means are provided for comparing a pixel cell group including the center pixel from the sampling window with a plurality of prediction bit subsets, certain ones of the prediction bit subsets being represented in a plurality of rotated orientations in the predicting means, and further for providing command signals indicating a match of the pixel cell group with one of the prediction bit subsets. Means responsive to a signaled match are provided for selecting the sampling window of pixels in accordance with the error signals associated with a matching prediction bit subset. Model means are provided for comparing the rotated sampling window of pixels with a plurality of stored sampling windows to determine the presence of a match of selected bits in the rotated sampling window with selected bits of a stored sampling window, correction means being provided for generating a signal indicative of an improved pixel representation for the center pixel in response to the determined match.

DESCRIPTION OF THE DRAWINGS

Fig. 1 is a schematic diagram showing a known window/model matching procedure used by known pixel image enhancement systems.

Fig. 2 is a block diagram illustrating the invention.

Fig. 3 shows the effect of the rotary module 64 in Fig. 2.

Figure 4 is an enlarged view of the prediction ROM/logic block 70 of Figure 2.

Fig. 5 shows exemplary input window configurations and the resulting window rotations thereof based on a comparison with a prediction bit set.

Figure 6 shows the plurality of prediction bit sets stored in the prediction ROM/logic, along with rotations thereof.

Fig. 7 is a logic flow diagram illustrating a method by which a reduced amount of Mu steroid comparison models for incorporation into a model ROM.

Figures 8 to 11 show exemplary model inputs that demonstrate how data reduction is achieved by operating the flow chart of Figure 7.

DETAILED DESCRIPTION OF THE INVENTION

2, portions of a laser printer 50 are shown that are involved in carrying out the pixel image enhancement method of the invention. A processor 52 controls the overall operation of the system, including the laser print engine 53, via communications on a bus 54. Connections between the bus 54 and the remaining portions of the system are only partially shown to avoid over-complicating the view. A random access memory (RAM) 56 stores a pixel bit map image to be enhanced in accordance with the invention. The bit map image is well known in the art and comprises a plurality of raster scan lines of ones and zeros that represent the individual pixel states in the bit map image.

A shift register network 58 is controlled by a processor 52 to move an input window sequentially across the bit map in the RAM 56. The circuitry comprising the shift register network 58 is well known in the art, a description of which can be found in U.S. Patents 4,847,641 and 4,437,122, cited above. Essentially, the shift register network 58 outputs a series of contiguous row segments of the raster bit map image stored in the RAM 56, hereinafter referred to as an "input window" (shown as 60 in FIG. 2).

The aim of the circuit of Fig. 2 is to determine determine whether a center pixel 62 in the input window 66 should remain unchanged or should be modified to better represent the pixel image present in the input window 60. During the course of operation of the circuit shown in Figure 2, the input window 60 is incremented throughout the bit map stored in the RAM 56 so that each bit position therein becomes a center pixel 62 and is subjected to the logical treatment described hereinafter.

The input window 60, as supplied by the shift register network 58, is provided to a rotation module 64. The rotation module 64 includes a 4:1 multiplexer for each pixel cell in the input window 60. Since the input window 60 (in this example) has sixty-one cells, the rotation module 64 includes sixty-one multiplexers, with each multiplexer having four inputs and one output. Each of the four inputs to a multiplexer represents one of four possible cell positions that a pixel can have when the input window 60 is rotated 0º (no rotation), 90º, 180º, or 270º. As shown in Figure 3, four cell positions in the input window 60 are connected as inputs to a multiplexer 65 in the rotation module 64. The bit state in each cell position is fed as an input to a multiplexer 65, and according to control inputs R1, R2, one of these bit states is passed as an output and becomes the bit state of a "rotated" window 66. By properly controlling each multiplexer, the input window 60 can thus be "rotated" by causing each of the multiplexers in the rotation module 64 (say, sixty-one in number) to select an input of a rotated cell position for connection to its output terminal.

The rotation module 64 is controlled by output signals from a prediction ROM/logic module 70, which in turn is supplied with input signals which determine the bit states of the middle nine cells of the input window 60. In Fig. 4, an enlarged view of the prediction module 70 is shown, with the center cell group 72 shown as an input thereto. The prediction module 70 includes a plurality of 3 x 3 prediction cell groups which, when encountered in an input center cell group of the window 60, indicate a high probability that a comparison model will be encountered in a model ROM/logic module 74, possibly in a rotated version (see Fig. 2).

A match between a prediction cell group and an input center cell group 72 in the prediction module 70 causes a command to be issued indicating whether the input window 60 should be rotated to match a model stored in the model module 74. This pre-rotation allows the number of models stored in the model module 74 to be significantly reduced from that taught in the prior art. When a match is found between the rotated window 66 and a model stored in the model module 74, a center pixel modification signal is provided which is subsequently rotated back in a rotate back module 76 to allow a proper center pixel modification signal to be generated and transmitted to the laser print engine 53.

As indicated above, the prediction module 70 includes a plurality of 3 x 3 prediction cell groups which, when matched with a center cell group from an input window 60, cause appropriate command signals to be issued to the rotation module 64. For example, when the center cell group 72 (Fig. 4) matches the prediction cell group 78, a high output signal appears on an output line 80 indicating that the input window 60 should be rotated 90° to match a model stored in the model module 74. This Output signal is applied to an OR circuit 82, causing output signal R1 to rise. Since center window 72 matches prediction cell group 78 (and as a result may not match prediction cell groups 84 or 86), output lines 88 and 90 remain low. Consequently, OR circuit 92 maintains R2 low, with the states of R1 and R2 forming command signals to rotate module 64.

The output signal levels manifested on R1 and R2 cause the spin module 64 to enable a second input signal to each multiplexer therein. A 90° rotation of the input window bit states in the input window 60 in a clockwise (CW) direction is the result. In contrast, if a match had been found between the center cell group 72 and the prediction cell group 84, the states of R1 and R2 would cause a third input signal to be output from each of the multiplexers in the spin module 64 on line 68, and so on. However, if the center cell group 72 matches the prediction cell group 94, no rotation takes place, with the multiplexers in the spin module 64 providing an unrotated output window on lines 68.

In Fig. 5, further examples of exemplary window input are shown in the four rotation states 96, 98, 100 and 102. The resulting center 3 x 3 cell groups are shown at 104, 106, 108 and 110. Assuming that the prediction cell group 94 (Fig. 4) is stored in the form of rotated versions 78, 84 and 86 in the prediction module 70, output signals R1, R2 will cause the window rotations shown at 112 in Fig. 5. In this case, it is important to realize that the upper right pixel position in the prediction cell group 94 (Fig. 4) is a "ignore" cell 114. Thus, the fact that the center cell groups 104, 106, 108 and 110 have a zero bit 116 (in the rotated states of these positions) is not taken into account. sition) is ignored in the comparison process. Thus, center cell group 104 exactly matches prediction cell group 78 and causes a 90º clockwise rotation output to be shown through R1 and R2. Similarly, a window such as 98 will cause no rotation to occur since it will not find a match in any of prediction cell groups 78, 84 or 86. Similarly, center cell group 108 will cause a 270º clockwise rotation output and center cell group 110 will cause a 180º clockwise rotation output.

In Fig. 6, prediction cell groups stored in the prediction module 70 are shown. They include 90º, 180º, and 270º counterclockwise (CCW) rotations of prediction cell groups 120. In addition, the prediction module 70 includes 90º counterclockwise rotations of prediction cell groups 122. The effect of the comparison actions in the prediction module 70 is to allow the input window 60 (Fig. 2) to be rotated to match a pre-stored model in the model module 74. Without the prediction module 70 and rotation module 64, the model module 74 would require up to four rotated copies of each model to allow a proper subpixel correction value to be output. Consequently, considerable memory is saved while allowing a very fast comparison to take place between an input window and the models stored in the model module 74.

Once a comparison is made, the model module 74 generates an output signal indicating an enhanced pixel representation to be applied to the center pixel 62 of the input window 60. Since the enhanced pixel representation may be rotated, it is fed to a reverse rotation module 76 where the input signals R1 and R2 allow a reverse rotation of the four possible output signals. from the model module 74 and cause a rotated back enhanced pixel representation to appear on one of the output lines 124. Essentially, the rotate back module 76 produces the exact complementary function that each of the multiplexers in the rotate module 64 produces. However, there are only four multiplexers in the rotate back module 76, whereas sixty-one exist in the rotate module 64 (assuming an input window 60 as shown).

The output signal levels appearing on lines 124 from the spin-back module 76 are fed to a controller in the laser print engine 53 which appropriately modulates the laser beam to produce the output sub-pixel array displayed. The specific modulation operation is described in U.S. Patent 4,847,641 issued to Tung, the assignee of the present application, the contents of which are hereby incorporated by reference.

Another aspect of the invention is the determination of the reduced model set contained in the model module 74 of Figure 2. The selection of which original models are included in the reduced model set and which models are removed is based on the nature of the original model set and the characteristics of the architecture of the invention. If an input window that matches a model of the original model set is present, it is rotated based on its mean cell group comparison with the predictor cell groups. In order to maintain a match for the rotated window, the original model must be rotated by the same amount and included in the reduced model set.

In Fig. 7, a method is shown that transforms an original set of pattern matching models into a reduced set stored in a model module 74. The method determines in advance all possible input window rotations for all original model set matches. The limited number of models, the small 3 x 3 prediction cell group size, and the limited number of prediction cell groups makes an exhaustive test possible.

Initially, an input model is accessed from a file (block 130) and all possible predicted rotations are determined (block 132) based on the model mean 3 x 3 cell group (which are determined in combination with the prediction cell groups in the prediction module 70). The first predicted rotation is accessed (block 134), and the accessed model is rotated by the predicted rotation (block 136). If the accessed model is ambiguous, a rotation (R1, R2) dependency is added to the rotated model, as explained below (block 138).

The current set of reduced models is searched to see if the rotated model has already been included (decision block 140). If it has been included, the rotated model is redundant and can be eliminated (block 142). If it has not been included, the rotated model is added to the current set of reduced models (block 144).

If additional predicted rotations exist for the accessed model that occur as a result of comparing them to prediction cell groups (decision block 146), the next predicted rotation is accessed (block 148) and the process returns to block 136. If no more predicted rotations exist, a determination is made as to whether there are any more models in the input file (decision block 150), and if so, the process returns. If no more models exist, the current set of reduced models is referred to as the reduced model set. stored for use in the model module 74.

Fig. 8 shows four members of the original model set, all rotated variations of the same elementary model. The first original model 160 has a disregard cell 162 in the upper right corner of the center cell group. Since an input window matching this model can have either a black or a white bit at this position, both possibilities for its predicted rotations are examined. If no possible center cell group matches a predicted cell group, input windows matching this model are not rotated. A 0º rotation of the model must be included in the reduced model set to match the input window.

The second original model 164 has an ignored cell 166 in the center cell group, and both possibilities are examined. Both possible center cell groups 168, 170 match a prediction cell group of 270º clockwise rotation, so the model is rotated 270º clockwise (since the same would be an input window match). It can be seen that the rotated model 172 is identical to that of the first original model 160. In a similar manner, the remaining two original models (174, 176) each have two possible center cell groups that are examined and are rotated (since matching input windows would exist). The rotated models of the same are also identical to those of the first original model 160. Only one occurrence of this rotated model 178 is required in the reduced model set.

Fig. 9 shows four members of the original model set, all rotated variations of the same elementary model, all with one disregardable bit in the center cell group. However, if both possible center cells groups are examined for the first original model 180, two different rotations 182, 184 are predicted. An input window with a white bit in the disregarded position will not be rotated, so a 0º rotation 186 of the original model 180 must be included in the reduced model set. However, an input window with a black bit in the disregarded position has a center cell group that matches a 270º clockwise rotation prediction cell group 184, and the window will be rotated 270º clockwise. Thus, a 270º clockwise rotation 188 of the original model must also be included in the reduced model set.

Examination of the possible center cell groups for the remaining three original models results in two different rotations for each, yielding a pair of rotated models identical to the pair found for the first original model. Only one occurrence of each model in the pair is needed in the reduced model set.

Fig. 10 shows an example of a member 190 of an original model set that has no rotated variations. Input windows that match this model are rotated 90° clockwise, so a 90° clockwise rotation 192 of the original model must be included. The first input window 194 shown in Fig. 11 matches the original model 190, with its rotated window 195 matching the rotated model 192. However, the second input window 196 does not match the original model 190, but with its rotated window 197 identical to that of the first input window 192. This would result in an erroneous model match. Original models such as these are said to be ambiguous.

Ambiguous models can be accommodated by adding a rotation dependency to the reduced model. If a rotated window, in addition to matching the white and black cells defined in the reduced model, must also have a specific rotation (state of R1 and R2 in Fig. 2) to match the reduced model, erroneous model matches can be eliminated. Adding a rotation dependency requires R1 and R2 inputs to the model module (not shown in Fig. 2). Ambiguous models are not reduced by this invention, although they are not excluded from it.

It should be understood that the foregoing description is merely illustrative of the invention. Various alternatives and modifications may be devised by those skilled in the art without departing from the invention. Thus, the present invention is intended to embrace all such alternatives, modifications and variations that fall within the scope of the appended claims.

Claims (9)

1. An apparatus for improving the reproduction of images represented by a plurality of pixels, the apparatus having the following features: a storage device (58) for storing successive lines of a bit map image; means for selecting a sampling window (60) of pixels from the bit map image in the storage device, the sampling window of pixels comprising a plurality of row segments of bits representing the pixels, the sampling window having a center bit representing a center pixel (62) in the sampling window; a predictor (70) for comparing a pixel cell group (72) containing the central pixel (62) from the sampling window with a plurality of prediction bit subsets (70, 78, 84, 86), certain ones of the prediction bit subsets being presented in a plurality of rotated orientations in the predictor, and further for providing command signals (R1, R2) indicating a match of the pixel cell group with one of the prediction bit subsets; means (64, 65) responsive to a signaled match for rotating (66) the sampling window of pixels in accordance with the command signals associated with the match prediction bit subset; a model device (74) for comparing the rotated th sampling window (66) of pixels with a plurality of stored sampling windows (178, 186, 188) to determine the existence of a match of selected bits in the rotated sampling window with selected bits of a stored sampling window; and a correction device (76) for generating a signal (124) indicative of an improved pixel representation for the central pixel in response to the determined match. 2. The apparatus of claim 1, wherein the means (64, 65) for rotating comprises switching means (4:1 MUX) for each pixel position in the sampling window, the switching means being operative to provide output signals indicative of a rotated sampling window of pixels in accordance with the command signals (R1, R2) from the predicting means (70). 3. The apparatus of claim 1 or 2, wherein the corrector (76) receives an input signal from the modeler (74) indicative of a partial pixel representation, the corrector rotating the partial pixel representation back to an orientation that matches the sampling window of pixels prior to the rotation. 4. The apparatus of claim 3, wherein the means (64, 65) for rotating and the correcting means (76) are responsive to identical command signals (R1, R2) from the predicting means (70) to achieve the rotating and reversing actions. 5. The apparatus of claim 3, wherein the reversed partial pixel representation is provided as an output signal for modulating a laser printing engine (53). 6. The apparatus of claim 5, wherein the correction means further comprises four switching means (4:1 MUX), each switching means connected to cause reversal of the four possible partial pixel representations from the model means and to provide a properly aligned partial pixel representation for controlling the laser printing engine (53). 7. The apparatus of any preceding claim, wherein the predictor (70) comprises a plurality of stored NxM prediction bit subsets, where N may be equal to or not equal to M, certain ones of the subsets appearing in a plurality of orthogonal orientations with respect to one another. 8. A method for improving the reproduction of images represented by a plurality of pixels, comprising the steps of: (a) storing (58) consecutive rows of a bitmap image; (b) selecting a sampling window (60) of initial alignment comprising a plurality of row segments of bits representing pixels from the bitmap image, the sampling window having a center pixel bit (62); (c) comparing (70) a pixel bit subset (72) from the sampling window containing the center pixel bit (62) with a plurality of prediction bit subsets (70, 78, 84, 86), wherein certain prediction bit subsets are represented in a plurality of rotated orientations, and issuing a command (R1, R2) that causes a match of the pi xelbit subset with a prediction bit subset when such a match is found; (d) rotating (64), if necessary, the selected sampling window according to the command (R1, R2) issued as a result of a found match with a prediction bit subset; (e) comparing (74) the rotated selected sample window (66) with a plurality of stored sample windows (178, 186, 188) to determine whether a match exists between selected bits in the rotated selected sample window and identical selected bits of a stored sample window; and (f) In response to a found match, substituting the central pixel with an enhanced pixel representation (124). 9. The method according to claim 8, wherein the comparing and substituting steps (e) and (f) comprise the following steps: displaying a subpixel representation corresponding to a found, matching, stored sample window; and Rotating back (76), if necessary, the subpixel representation to match the initial orientation of the selected sampling window (60).